Changelog

PR Workshop

• New PR Workshop tab: Browse all Arbiter-created PRs (open, closed, and merged) in one place
• AI chat per PR: Ask questions about a PR's changes, get explanations of the diff, or request follow-up work — all in a persistent conversation tied to that PR
• Full context awareness: The AI has access to the PR diff, changed files, original issue, and commit history when answering
• Search and filter: Find PRs quickly by title, number, branch, or state
• Persistent history: Conversation history is saved per PR so you can pick up where you left off

Analytics: Merged PRs by Worker

• New chart on the analytics dashboard showing how many PRs each worker has gotten merged, broken down by worker type
• Scoped to the selected time range, filtered by when Arbiter opened the PR — consistent with all other dashboard metrics
• PR state is synced from GitHub in the background and cached locally, so the dashboard loads instantly regardless of how many PRs you have
• Sync automatically runs on startup and every 15 minutes; refresh button and time range changes trigger a fresh sync

Worker Error Handling

• Errors now surface in the UI instead of silently disappearing into the terminal — worker failures show as a toast notification and update the run status
• Fixed a crash where a missing worker binary (e.g. Claude Code or Cursor Agent not on PATH) would throw an unhandled exception and leave the UI stuck at "Running..."
• Error messages are now human-readable: a missing binary tells you exactly what isn't installed and how to fix it, and CLI errors surface the actual message from the tool rather than a generic exit code
• Timeouts are handled separately from hard failures — timed-out runs still check for partial work, while genuine errors fail fast and clean up

Bug Fixes

• Fixed Elasticsearch integration failing in compiled binaries due to Bun compatibility issues with the @elastic/elasticsearch client

Version 1.0 is here! After extensive testing and community feedback, Arbiter is officially stable and ready for production use.

What's New

• Minor UI polish and improvements throughout the dashboard
• Updated pricing structure with launch week sale
• Various stability improvements and bug fixes

Thank you! To everyone who tested, provided feedback, and helped shape Arbiter into what it is today. Here's to v1.0 and beyond!

New Features

• Browser now automatically opens when launching the dashboard with arbiter ui
• CLI now checks for updates on startup and notifies you when a new version is available
• Added option to manually enter a license key directly from the trial badge in the dashboard

Improvements

• Renamed license badge from "PRO" to "LICENSED" for clarity

Bug Fixes

• Fixed broken social media links in footer

Improvements

• License expiration overlay now shows appropriate messaging for both trial and license expiration
• Sanitized placeholder text in UI configuration fields

Bug Fixes

• Fixed an issue with npm installations

First stable release! Arbiter v0.1.1 brings major improvements to observability, AI capabilities, and developer experience. This release includes extensive new features based on community feedback.

New AI Models

• Latest Claude Models: Support for Opus 4.6, Sonnet 4.5, and Haiku 4.5
• Codex Integration: Support for Codex AI models
• Local Model Support: Run Ollama models locally (llama3.1, qwen2.5-coder, deepseek-coder)
• LM Studio Integration: User-friendly GUI for local model management
• OpenCode CLI: Open-source CLI with multi-provider support

Dashboard & Observability

• Analytics Dashboard: Comprehensive metrics with run history charts and worker breakdown
• Workspace Viewer: Real-time monitoring of active workspaces with live logs
• Jump-In Feature: Open terminals directly into running workspaces for debugging
• Run History Table: Paginated view of all runs with filtering and search
• Visual Stats: Success rates, average runtime, and performance metrics

Self-Learning Mode

• Automatic Knowledge Extraction: AI learns from successful runs and mistakes
• Project-Specific Learnings: Builds custom knowledge base per repository
• Learning Categories: Code patterns, gotchas, testing requirements, and edge cases
• Persistent Memory: New runs benefit from accumulated knowledge
• Manual Learning Management: Edit, view, and delete learnings through the dashboard

Webhooks & Notifications

• Custom Webhooks: Create HTTP endpoints for event notifications
• Rich Events: PR opened, run completed/failed, validation/checks failed, AI review completed
• HMAC Signatures: Verify webhook authenticity with secret signing
• Event Filtering: Subscribe only to events you care about
• Integration Ready: Works with Slack, Discord, PagerDuty, and custom services

Advanced Logging

• PostgreSQL Support: Persistent log storage with full-text search
• Elasticsearch Integration: Scalable log indexing and querying
• Structured Logging: JSON-formatted logs for easy parsing and analysis
• Dashboard Log Viewer: Browse and search logs directly in the UI

PR/Issue Commands

• /arbiter-change: Request changes to Arbiter-created PRs with natural language
• /arbiter-review: Trigger manual AI code review on any PR
• #arbiter-note: Add context and notes for AI to consider during work
• Write Access Required: Authorization checks prevent abuse in public repos
• Priority Processing: Comment-triggered work takes precedence over new issues

Multiple Installation Methods

• npm Package: Install globally with npm install -g @git-arbiter/arbiter
• Homebrew: Easy installation on macOS with brew install git-arbiter/tap/arbiter
• Docker: Pre-configured Docker images for containerized deployment
• Direct Download: Pre-built binaries for all major platforms (macOS, Linux, ARM64)

Workspace Management

• CLI Commands: arbiter validate, arbiter commit, arbiter push, arbiter pr
• Manual Intervention: Jump into workspaces to fix issues manually
• Workspace Deletion: Remove completed workspaces from the dashboard
• Path Copying: Quick access to workspace directories

AI Code Review

• Automatic Reviews: AI reviews all Arbiter-created PRs before opening
• Manual Triggers: Use /arbiter-review on any PR
• Custom Prompts: Configure review criteria per repository
• Verdict System: APPROVE, REQUEST_CHANGES, or COMMENT outcomes
• Detailed Feedback: Security issues, code quality, and improvement suggestions

Configuration Improvements

• Model Discovery: Auto-detect available Ollama/LM Studio models
• Validation Pipeline: Customizable test/lint/build commands per repo
• Auto-Run Settings: Configurable intervals and PR iteration priority
• Security Policies: Granular command allowlists and blocklists
• Enhanced Dashboard: Real-time status updates and configuration viewing

Ready for Production: v0.1.1 is our first stable release suitable for production use. All beta features have been thoroughly tested and hardened. Thank you to our beta testers for invaluable feedback!

Initial beta release! Arbiter is now available for early adopters. This release includes all core functionality for autonomous GitHub issue solving.

Core Features

• Autonomous GitHub issue monitoring and resolution
• Multi-AI support: Claude Code, Cursor, Ollama, and LM Studio
• Automatic pull request creation and management
• PR iteration support with comment-based commands
• Web dashboard for real-time monitoring
• Local-first architecture - your code never leaves your machine

CLI & Workflow

• Simple label-based issue assignment (e.g., agent-ok)
• Automated workspace setup and cleanup
• Git workflow automation (branches, commits, PRs)
• Test execution and validation
• Security layer for sensitive commands

Deployment

• Docker-based deployment for easy setup
• Support for multiple deployment modes (host CLI, host network)
• Environment variable configuration
• Volume mounting for local code access

Security & Privacy

• Local-first processing - no external code hosting
• GitHub PAT authentication
• Configurable security policies
• Protected command execution

What's Next

• Enhanced error handling and recovery
• Additional AI provider integrations
• Improved dashboard features
• Community feedback and feature requests

Beta Notice: This is an early release. We're actively working on improvements and welcome your feedback. Join our Discord community to report bugs, request features, and get support.